Among the masses of generative AI startups that have popped up all over Europe in the past two years — global VC firm Accel counts just under 200 in the region — there is one rare breed: profitable ones.
Paris-based startup Photoroom, a Y Combinator alumni that provides a tool powered by GenAI for photo editing, is one of them.
Founded in 2019 and now valued at €500m, Photoroom broke even a year after it launched, and is still profitable. It reached €50m in annual recurring revenue last year; cofounder Matthieu Rouif says the numbers have kept growing since, but declined to provide an updated figure.
Photoroom raised a $43m Series B in February, which Rouif says came on top of the company already "self-funding" its activities — and indicating that it now has significant amounts of cash in the bank to deploy.
Profitability sets the company apart from many of its peers. In Europe, few startups in the sector — including Germany’s Aleph Alpha, France’s Mistral AI and the UK’s Stability AI — have publicly reported their annual revenues, let alone a net profit.
One exception is Paris-based startup Animaj, which leverages GenAI to create video content for children. The company previously reported to Sifted an EBITDA of “a few million euros”, with annual revenues surpassing €10m.
OpenAI, one of the hottest companies in the space, is reported to be facing an operational loss of $5bn this year.
“[Being a profitable GenAI startup] is pretty unique worldwide,” says Rouif. “At our level of revenue, it’s probably less than a handful of us.”
Building Photoroom
Rouif created Photoroom with cofounder Eliot Andres five years ago, with a vision to put AI-powered tools into the hands of small businesses to help them create high-quality visuals when advertising their products online.
“If you’re a shopkeeper or a restaurant, today your shopfront is the internet,” says Rouif. “Without good pictures, you’re dead.”
Via the Photoroom app, users can easily remove and replace backgrounds — and produce visuals for their products that resemble studio-shot pictures, without the need for a large marketing budget.
The startup started looking at GenAI technologies a couple of years ago, as it searched for ways to enable users to prompt the app with requests for specific scenes they wanted to see their products staged in.
Existing image-generating tools like OpenAI’s Dall-E were not capable of delivering this, says Rouif, so the founders decided to build their own GenAI model. The model launched in the app at the beginning of this year, after a year and a half of training.
The numbers show the technology is resonating with users: Rouif says Photoroom is nearing 200m downloads to date, with customers telling him the app is enabling notable improvements in their business.
Over the past few months, digital marketing platform Smartly used the Photoroom app to create visuals for sleep products as part of an advertising campaign; it said that it saw a 72% increase in click-through rates and a 286% increase in average order value.
“Even if Photoroom is not very expensive — at about €9 a month — the best way to prove that the app is useful is to see that people are paying,” says Rouif. “Users vote with their credit card.”
The app offers a freemium plan and several premium options. Rouif declined to disclose the conversion rate to premium plans.
How did Photoroom become profitable?
Building a GenAI tool does not come for cheap. Recent months have seen record-breaking fundraises from companies in the space, such as Poolside’s €500m fundraise, Mistral’s €600m round, and Aleph Alpha’s €460m Series B — numbers that reflect the huge compute costs of training the models.
Photoroom trains image-based models that are less costly to develop than the text-based models developed by Mistral and Aleph Alpha. The startup has so far raised $64m.
A lower GPU bill may partly explain why the startup managed to reach profitability in such a short time. But Rouif says Photoroom’s secret sauce mostly comes down to the company always being built to answer an existing user need.
“Our approach has been to start with users’ needs to develop useful models,” says Rouif. “The secondary effect of this is to create revenue, because you know exactly what the purpose of your product is.”
With this, Photoroom has been able to attract a team of 80 employees spread out across Europe, with a large proportion based in Paris. In the French capital, the startup is known to be actively seeking out top AI talent — and to be ready to hand out generous pay packages to secure the best candidates.
“Salaries for machine learning roles have increased by 50% in a year for us,” says Rouif. “We’re very happy to be able to be ambitious [in recruitment].”
GenAI’s profitability equation
Despite receiving plenty of VC attention in recent years, the majority of GenAI startups have not yet been able to find profitable business models.
Some investors say this is partly due to these companies not finding the right user needs to serve.
In an article published last year, Sequoia Capital investors Sonya Huang and Pat Grady wrote: “A whisper [has begun] to spread within Silicon Valley that GenAI [is] not actually useful.
“Early signs of success don’t change the reality that a lot of [Gen] AI companies simply do not have product-market fit or a sustainable competitive advantage.”
Rouif says: “There will be some time before the revenues [of GenAI startups] catch up with the investments. There will be several waves, a bit like the internet, but I think it will come.
“Of course at some point, lots of startups in the ecosystem will also shut down.”
Even though some GenAI startups will find it harder to monetise their business models — such as open source, an approach taken by Mistral among others — Rouif thinks all these approaches are important for the development of the sector.
“Without open source, we wouldn’t exist,” says Rouif. “So I’m never going to say that others aren’t doing it right.’